1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 5 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org> 6 * Copyright (c) 2009 Apple, Inc. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 #include <sys/cdefs.h> 32 #include "opt_ktrace.h" 33 #include "opt_kqueue.h" 34 35 #ifdef COMPAT_FREEBSD11 36 #define _WANT_FREEBSD11_KEVENT 37 #endif 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/capsicum.h> 42 #include <sys/kernel.h> 43 #include <sys/limits.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/proc.h> 47 #include <sys/malloc.h> 48 #include <sys/unistd.h> 49 #include <sys/file.h> 50 #include <sys/filedesc.h> 51 #include <sys/filio.h> 52 #include <sys/fcntl.h> 53 #include <sys/jail.h> 54 #include <sys/jaildesc.h> 55 #include <sys/kthread.h> 56 #include <sys/selinfo.h> 57 #include <sys/queue.h> 58 #include <sys/event.h> 59 #include <sys/eventvar.h> 60 #include <sys/poll.h> 61 #include <sys/protosw.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sbuf.h> 64 #include <sys/sigio.h> 65 #include <sys/signalvar.h> 66 #include <sys/socket.h> 67 #include <sys/socketvar.h> 68 #include <sys/stat.h> 69 #include <sys/sysctl.h> 70 #include <sys/sysent.h> 71 #include <sys/sysproto.h> 72 #include <sys/syscallsubr.h> 73 #include <sys/taskqueue.h> 74 #include <sys/uio.h> 75 #include <sys/user.h> 76 #ifdef KTRACE 77 #include <sys/ktrace.h> 78 #endif 79 #include <machine/atomic.h> 80 #ifdef COMPAT_FREEBSD32 81 #include <compat/freebsd32/freebsd32.h> 82 #include <compat/freebsd32/freebsd32_util.h> 83 #endif 84 85 #include <vm/uma.h> 86 87 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 88 89 /* 90 * This lock is used if multiple kq locks are required. This possibly 91 * should be made into a per proc lock. 92 */ 93 static struct mtx kq_global; 94 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 95 #define KQ_GLOBAL_LOCK(lck, haslck) do { \ 96 if (!haslck) \ 97 mtx_lock(lck); \ 98 haslck = 1; \ 99 } while (0) 100 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 101 if (haslck) \ 102 mtx_unlock(lck); \ 103 haslck = 0; \ 104 } while (0) 105 106 TASKQUEUE_DEFINE_THREAD(kqueue_ctx); 107 108 static int kevent_copyout(void *arg, struct kevent *kevp, int count); 109 static int kevent_copyin(void *arg, struct kevent *kevp, int count); 110 static int kqueue_register(struct kqueue *kq, struct kevent *kev, 111 struct thread *td, int mflag); 112 static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 113 static void kqueue_release(struct kqueue *kq, int locked); 114 static void kqueue_destroy(struct kqueue *kq); 115 static void kqueue_drain(struct kqueue *kq, struct thread *td); 116 static int kqueue_expand(struct kqueue *kq, const struct filterops *fops, 117 uintptr_t ident, int mflag); 118 static void kqueue_task(void *arg, int pending); 119 static int kqueue_scan(struct kqueue *kq, int maxevents, 120 struct kevent_copyops *k_ops, 121 const struct timespec *timeout, 122 struct kevent *keva, struct thread *td); 123 static void kqueue_wakeup(struct kqueue *kq); 124 static const struct filterops *kqueue_fo_find(int filt); 125 static void kqueue_fo_release(int filt); 126 struct g_kevent_args; 127 static int kern_kevent_generic(struct thread *td, 128 struct g_kevent_args *uap, 129 struct kevent_copyops *k_ops, const char *struct_name); 130 131 static fo_ioctl_t kqueue_ioctl; 132 static fo_poll_t kqueue_poll; 133 static fo_kqfilter_t kqueue_kqfilter; 134 static fo_stat_t kqueue_stat; 135 static fo_close_t kqueue_close; 136 static fo_fill_kinfo_t kqueue_fill_kinfo; 137 138 static const struct fileops kqueueops = { 139 .fo_read = invfo_rdwr, 140 .fo_write = invfo_rdwr, 141 .fo_truncate = invfo_truncate, 142 .fo_ioctl = kqueue_ioctl, 143 .fo_poll = kqueue_poll, 144 .fo_kqfilter = kqueue_kqfilter, 145 .fo_stat = kqueue_stat, 146 .fo_close = kqueue_close, 147 .fo_chmod = invfo_chmod, 148 .fo_chown = invfo_chown, 149 .fo_sendfile = invfo_sendfile, 150 .fo_cmp = file_kcmp_generic, 151 .fo_fill_kinfo = kqueue_fill_kinfo, 152 }; 153 154 static int knote_attach(struct knote *kn, struct kqueue *kq); 155 static void knote_drop(struct knote *kn, struct thread *td); 156 static void knote_drop_detached(struct knote *kn, struct thread *td); 157 static void knote_enqueue(struct knote *kn); 158 static void knote_dequeue(struct knote *kn); 159 static void knote_init(void); 160 static struct knote *knote_alloc(int mflag); 161 static void knote_free(struct knote *kn); 162 163 static void filt_kqdetach(struct knote *kn); 164 static int filt_kqueue(struct knote *kn, long hint); 165 static int filt_procattach(struct knote *kn); 166 static void filt_procdetach(struct knote *kn); 167 static int filt_proc(struct knote *kn, long hint); 168 static int filt_jailattach(struct knote *kn); 169 static void filt_jaildetach(struct knote *kn); 170 static int filt_jail(struct knote *kn, long hint); 171 static int filt_fileattach(struct knote *kn); 172 static void filt_timerexpire(void *knx); 173 static void filt_timerexpire_l(struct knote *kn, bool proc_locked); 174 static int filt_timerattach(struct knote *kn); 175 static void filt_timerdetach(struct knote *kn); 176 static void filt_timerstart(struct knote *kn, sbintime_t to); 177 static void filt_timertouch(struct knote *kn, struct kevent *kev, 178 u_long type); 179 static int filt_timervalidate(struct knote *kn, sbintime_t *to); 180 static int filt_timer(struct knote *kn, long hint); 181 static int filt_userattach(struct knote *kn); 182 static void filt_userdetach(struct knote *kn); 183 static int filt_user(struct knote *kn, long hint); 184 static void filt_usertouch(struct knote *kn, struct kevent *kev, 185 u_long type); 186 187 static const struct filterops file_filtops = { 188 .f_isfd = 1, 189 .f_attach = filt_fileattach, 190 }; 191 static const struct filterops kqread_filtops = { 192 .f_isfd = 1, 193 .f_detach = filt_kqdetach, 194 .f_event = filt_kqueue, 195 }; 196 /* XXX - move to kern_proc.c? */ 197 static const struct filterops proc_filtops = { 198 .f_isfd = 0, 199 .f_attach = filt_procattach, 200 .f_detach = filt_procdetach, 201 .f_event = filt_proc, 202 }; 203 static const struct filterops jail_filtops = { 204 .f_isfd = 0, 205 .f_attach = filt_jailattach, 206 .f_detach = filt_jaildetach, 207 .f_event = filt_jail, 208 }; 209 static const struct filterops timer_filtops = { 210 .f_isfd = 0, 211 .f_attach = filt_timerattach, 212 .f_detach = filt_timerdetach, 213 .f_event = filt_timer, 214 .f_touch = filt_timertouch, 215 }; 216 static const struct filterops user_filtops = { 217 .f_attach = filt_userattach, 218 .f_detach = filt_userdetach, 219 .f_event = filt_user, 220 .f_touch = filt_usertouch, 221 }; 222 223 static uma_zone_t knote_zone; 224 static unsigned int __exclusive_cache_line kq_ncallouts; 225 static unsigned int kq_calloutmax = 4 * 1024; 226 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 227 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 228 229 /* XXX - ensure not influx ? */ 230 #define KNOTE_ACTIVATE(kn, islock) do { \ 231 if ((islock)) \ 232 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 233 else \ 234 KQ_LOCK((kn)->kn_kq); \ 235 (kn)->kn_status |= KN_ACTIVE; \ 236 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 237 knote_enqueue((kn)); \ 238 if (!(islock)) \ 239 KQ_UNLOCK((kn)->kn_kq); \ 240 } while (0) 241 #define KQ_LOCK(kq) do { \ 242 mtx_lock(&(kq)->kq_lock); \ 243 } while (0) 244 #define KQ_FLUX_WAKEUP(kq) do { \ 245 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 246 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 247 wakeup((kq)); \ 248 } \ 249 } while (0) 250 #define KQ_UNLOCK_FLUX(kq) do { \ 251 KQ_FLUX_WAKEUP(kq); \ 252 mtx_unlock(&(kq)->kq_lock); \ 253 } while (0) 254 #define KQ_UNLOCK(kq) do { \ 255 mtx_unlock(&(kq)->kq_lock); \ 256 } while (0) 257 #define KQ_OWNED(kq) do { \ 258 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 259 } while (0) 260 #define KQ_NOTOWNED(kq) do { \ 261 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 262 } while (0) 263 264 static struct knlist * 265 kn_list_lock(struct knote *kn) 266 { 267 struct knlist *knl; 268 269 knl = kn->kn_knlist; 270 if (knl != NULL) 271 knl->kl_lock(knl->kl_lockarg); 272 return (knl); 273 } 274 275 static void 276 kn_list_unlock(struct knlist *knl) 277 { 278 bool do_free; 279 280 if (knl == NULL) 281 return; 282 do_free = knl->kl_autodestroy && knlist_empty(knl); 283 knl->kl_unlock(knl->kl_lockarg); 284 if (do_free) { 285 knlist_destroy(knl); 286 free(knl, M_KQUEUE); 287 } 288 } 289 290 static bool 291 kn_in_flux(struct knote *kn) 292 { 293 294 return (kn->kn_influx > 0); 295 } 296 297 static void 298 kn_enter_flux(struct knote *kn) 299 { 300 301 KQ_OWNED(kn->kn_kq); 302 MPASS(kn->kn_influx < INT_MAX); 303 kn->kn_influx++; 304 } 305 306 static bool 307 kn_leave_flux(struct knote *kn) 308 { 309 310 KQ_OWNED(kn->kn_kq); 311 MPASS(kn->kn_influx > 0); 312 kn->kn_influx--; 313 return (kn->kn_influx == 0); 314 } 315 316 #define KNL_ASSERT_LOCK(knl, islocked) do { \ 317 if (islocked) \ 318 KNL_ASSERT_LOCKED(knl); \ 319 else \ 320 KNL_ASSERT_UNLOCKED(knl); \ 321 } while (0) 322 #ifdef INVARIANTS 323 #define KNL_ASSERT_LOCKED(knl) do { \ 324 knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED); \ 325 } while (0) 326 #define KNL_ASSERT_UNLOCKED(knl) do { \ 327 knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED); \ 328 } while (0) 329 #else /* !INVARIANTS */ 330 #define KNL_ASSERT_LOCKED(knl) do {} while (0) 331 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 332 #endif /* INVARIANTS */ 333 334 #ifndef KN_HASHSIZE 335 #define KN_HASHSIZE 64 /* XXX should be tunable */ 336 #endif 337 338 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 339 340 static int 341 filt_nullattach(struct knote *kn) 342 { 343 344 return (ENXIO); 345 }; 346 347 static const struct filterops null_filtops = { 348 .f_isfd = 0, 349 .f_attach = filt_nullattach, 350 }; 351 352 /* XXX - make SYSINIT to add these, and move into respective modules. */ 353 extern const struct filterops sig_filtops; 354 extern const struct filterops fs_filtops; 355 356 /* 357 * Table for all system-defined filters. 358 */ 359 static struct mtx filterops_lock; 360 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF); 361 static struct { 362 const struct filterops *for_fop; 363 int for_nolock; 364 int for_refcnt; 365 } sysfilt_ops[EVFILT_SYSCOUNT] = { 366 [~EVFILT_READ] = { &file_filtops, 1 }, 367 [~EVFILT_WRITE] = { &file_filtops, 1 }, 368 [~EVFILT_AIO] = { &null_filtops }, 369 [~EVFILT_VNODE] = { &file_filtops, 1 }, 370 [~EVFILT_PROC] = { &proc_filtops, 1 }, 371 [~EVFILT_SIGNAL] = { &sig_filtops, 1 }, 372 [~EVFILT_TIMER] = { &timer_filtops, 1 }, 373 [~EVFILT_PROCDESC] = { &file_filtops, 1 }, 374 [~EVFILT_FS] = { &fs_filtops, 1 }, 375 [~EVFILT_LIO] = { &null_filtops }, 376 [~EVFILT_USER] = { &user_filtops, 1 }, 377 [~EVFILT_SENDFILE] = { &null_filtops }, 378 [~EVFILT_EMPTY] = { &file_filtops, 1 }, 379 [~EVFILT_JAIL] = { &jail_filtops, 1 }, 380 [~EVFILT_JAILDESC] = { &file_filtops, 1 }, 381 }; 382 383 /* 384 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 385 * method. 386 */ 387 static int 388 filt_fileattach(struct knote *kn) 389 { 390 391 return (fo_kqfilter(kn->kn_fp, kn)); 392 } 393 394 /*ARGSUSED*/ 395 static int 396 kqueue_kqfilter(struct file *fp, struct knote *kn) 397 { 398 struct kqueue *kq = kn->kn_fp->f_data; 399 400 if (kn->kn_filter != EVFILT_READ) 401 return (EINVAL); 402 403 kn->kn_status |= KN_KQUEUE; 404 kn->kn_fop = &kqread_filtops; 405 knlist_add(&kq->kq_sel.si_note, kn, 0); 406 407 return (0); 408 } 409 410 static void 411 filt_kqdetach(struct knote *kn) 412 { 413 struct kqueue *kq = kn->kn_fp->f_data; 414 415 knlist_remove(&kq->kq_sel.si_note, kn, 0); 416 } 417 418 /*ARGSUSED*/ 419 static int 420 filt_kqueue(struct knote *kn, long hint) 421 { 422 struct kqueue *kq = kn->kn_fp->f_data; 423 424 kn->kn_data = kq->kq_count; 425 return (kn->kn_data > 0); 426 } 427 428 /* XXX - move to kern_proc.c? */ 429 static int 430 filt_procattach(struct knote *kn) 431 { 432 struct proc *p; 433 int error; 434 bool exiting, immediate; 435 436 exiting = immediate = false; 437 if (kn->kn_sfflags & NOTE_EXIT) 438 p = pfind_any(kn->kn_id); 439 else 440 p = pfind(kn->kn_id); 441 if (p == NULL) 442 return (ESRCH); 443 if (p->p_flag & P_WEXIT) 444 exiting = true; 445 446 if ((error = p_cansee(curthread, p))) { 447 PROC_UNLOCK(p); 448 return (error); 449 } 450 451 kn->kn_ptr.p_proc = p; 452 kn->kn_flags |= EV_CLEAR; /* automatically set */ 453 454 /* 455 * Internal flag indicating registration done by kernel for the 456 * purposes of getting a NOTE_CHILD notification. 457 */ 458 if (kn->kn_flags & EV_FLAG2) { 459 kn->kn_flags &= ~EV_FLAG2; 460 kn->kn_data = kn->kn_sdata; /* ppid */ 461 kn->kn_fflags = NOTE_CHILD; 462 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); 463 immediate = true; /* Force immediate activation of child note. */ 464 } 465 /* 466 * Internal flag indicating registration done by kernel (for other than 467 * NOTE_CHILD). 468 */ 469 if (kn->kn_flags & EV_FLAG1) { 470 kn->kn_flags &= ~EV_FLAG1; 471 } 472 473 knlist_add(p->p_klist, kn, 1); 474 475 /* 476 * Immediately activate any child notes or, in the case of a zombie 477 * target process, exit notes. The latter is necessary to handle the 478 * case where the target process, e.g. a child, dies before the kevent 479 * is registered. 480 */ 481 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) 482 KNOTE_ACTIVATE(kn, 0); 483 484 PROC_UNLOCK(p); 485 486 return (0); 487 } 488 489 /* 490 * The knote may be attached to a different process, which may exit, 491 * leaving nothing for the knote to be attached to. So when the process 492 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 493 * it will be deleted when read out. However, as part of the knote deletion, 494 * this routine is called, so a check is needed to avoid actually performing 495 * a detach, because the original process does not exist any more. 496 */ 497 /* XXX - move to kern_proc.c? */ 498 static void 499 filt_procdetach(struct knote *kn) 500 { 501 502 knlist_remove(kn->kn_knlist, kn, 0); 503 kn->kn_ptr.p_proc = NULL; 504 } 505 506 /* XXX - move to kern_proc.c? */ 507 static int 508 filt_proc(struct knote *kn, long hint) 509 { 510 struct proc *p; 511 u_int event; 512 513 p = kn->kn_ptr.p_proc; 514 if (p == NULL) /* already activated, from attach filter */ 515 return (0); 516 517 /* Mask off extra data. */ 518 event = (u_int)hint & NOTE_PCTRLMASK; 519 520 /* If the user is interested in this event, record it. */ 521 if (kn->kn_sfflags & event) 522 kn->kn_fflags |= event; 523 524 /* Process is gone, so flag the event as finished. */ 525 if (event == NOTE_EXIT) { 526 kn->kn_flags |= EV_EOF | EV_ONESHOT; 527 kn->kn_ptr.p_proc = NULL; 528 if (kn->kn_fflags & NOTE_EXIT) 529 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); 530 if (kn->kn_fflags == 0) 531 kn->kn_flags |= EV_DROP; 532 return (1); 533 } 534 535 return (kn->kn_fflags != 0); 536 } 537 538 /* 539 * Called when the process forked. It mostly does the same as the 540 * knote(), activating all knotes registered to be activated when the 541 * process forked. Additionally, for each knote attached to the 542 * parent, check whether user wants to track the new process. If so 543 * attach a new knote to it, and immediately report an event with the 544 * child's pid. 545 */ 546 void 547 knote_fork(struct knlist *list, int pid) 548 { 549 struct kqueue *kq; 550 struct knote *kn; 551 struct kevent kev; 552 int error; 553 554 MPASS(list != NULL); 555 KNL_ASSERT_LOCKED(list); 556 if (SLIST_EMPTY(&list->kl_list)) 557 return; 558 559 memset(&kev, 0, sizeof(kev)); 560 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 561 kq = kn->kn_kq; 562 KQ_LOCK(kq); 563 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 564 KQ_UNLOCK(kq); 565 continue; 566 } 567 568 /* 569 * The same as knote(), activate the event. 570 */ 571 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 572 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 573 KNOTE_ACTIVATE(kn, 1); 574 KQ_UNLOCK(kq); 575 continue; 576 } 577 578 /* 579 * The NOTE_TRACK case. In addition to the activation 580 * of the event, we need to register new events to 581 * track the child. Drop the locks in preparation for 582 * the call to kqueue_register(). 583 */ 584 kn_enter_flux(kn); 585 KQ_UNLOCK(kq); 586 list->kl_unlock(list->kl_lockarg); 587 588 /* 589 * Activate existing knote and register tracking knotes with 590 * new process. 591 * 592 * First register a knote to get just the child notice. This 593 * must be a separate note from a potential NOTE_EXIT 594 * notification since both NOTE_CHILD and NOTE_EXIT are defined 595 * to use the data field (in conflicting ways). 596 */ 597 kev.ident = pid; 598 kev.filter = kn->kn_filter; 599 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | 600 EV_FLAG2; 601 kev.fflags = kn->kn_sfflags; 602 kev.data = kn->kn_id; /* parent */ 603 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 604 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 605 if (error) 606 kn->kn_fflags |= NOTE_TRACKERR; 607 608 /* 609 * Then register another knote to track other potential events 610 * from the new process. 611 */ 612 kev.ident = pid; 613 kev.filter = kn->kn_filter; 614 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 615 kev.fflags = kn->kn_sfflags; 616 kev.data = kn->kn_id; /* parent */ 617 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 618 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 619 if (error) 620 kn->kn_fflags |= NOTE_TRACKERR; 621 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 622 KNOTE_ACTIVATE(kn, 0); 623 list->kl_lock(list->kl_lockarg); 624 KQ_LOCK(kq); 625 kn_leave_flux(kn); 626 KQ_UNLOCK_FLUX(kq); 627 } 628 } 629 630 int 631 filt_jailattach(struct knote *kn) 632 { 633 struct prison *pr; 634 635 if (kn->kn_id == 0) { 636 /* Let jid=0 watch the current prison (including prison0). */ 637 pr = curthread->td_ucred->cr_prison; 638 mtx_lock(&pr->pr_mtx); 639 } else { 640 sx_slock(&allprison_lock); 641 pr = prison_find_child(curthread->td_ucred->cr_prison, 642 kn->kn_id); 643 sx_sunlock(&allprison_lock); 644 if (pr == NULL) 645 return (ENOENT); 646 if (!prison_isalive(pr)) { 647 mtx_unlock(&pr->pr_mtx); 648 return (ENOENT); 649 } 650 } 651 kn->kn_ptr.p_prison = pr; 652 kn->kn_flags |= EV_CLEAR; 653 knlist_add(pr->pr_klist, kn, 1); 654 mtx_unlock(&pr->pr_mtx); 655 return (0); 656 } 657 658 void 659 filt_jaildetach(struct knote *kn) 660 { 661 if (kn->kn_ptr.p_prison != NULL) { 662 knlist_remove(kn->kn_knlist, kn, 0); 663 kn->kn_ptr.p_prison = NULL; 664 } else 665 kn->kn_status |= KN_DETACHED; 666 } 667 668 int 669 filt_jail(struct knote *kn, long hint) 670 { 671 struct prison *pr; 672 u_int event; 673 674 pr = kn->kn_ptr.p_prison; 675 if (pr == NULL) /* already activated, from attach filter */ 676 return (0); 677 678 /* 679 * Mask off extra data. In the NOTE_JAIL_CHILD case, that's 680 * everything except the NOTE_JAIL_CHILD bit itself, since a 681 * JID is any positive integer. 682 */ 683 event = ((u_int)hint & NOTE_JAIL_CHILD) ? NOTE_JAIL_CHILD : 684 (u_int)hint & NOTE_JAIL_CTRLMASK; 685 686 /* If the user is interested in this event, record it. */ 687 if (kn->kn_sfflags & event) { 688 kn->kn_fflags |= event; 689 /* Report the created jail id or attached process id. */ 690 if (event == NOTE_JAIL_CHILD || event == NOTE_JAIL_ATTACH) { 691 if (kn->kn_data != 0) 692 kn->kn_fflags |= NOTE_JAIL_MULTI; 693 kn->kn_data = (kn->kn_fflags & NOTE_JAIL_MULTI) ? 0U : 694 (u_int)hint & ~event; 695 } 696 } 697 698 /* Prison is gone, so flag the event as finished. */ 699 if (event == NOTE_JAIL_REMOVE) { 700 kn->kn_flags |= EV_EOF | EV_ONESHOT; 701 kn->kn_ptr.p_prison = NULL; 702 if (kn->kn_fflags == 0) 703 kn->kn_flags |= EV_DROP; 704 return (1); 705 } 706 707 return (kn->kn_fflags != 0); 708 } 709 710 /* 711 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 712 * interval timer support code. 713 */ 714 715 #define NOTE_TIMER_PRECMASK \ 716 (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS) 717 718 static sbintime_t 719 timer2sbintime(int64_t data, int flags) 720 { 721 int64_t secs; 722 723 /* 724 * Macros for converting to the fractional second portion of an 725 * sbintime_t using 64bit multiplication to improve precision. 726 */ 727 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) 728 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) 729 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) 730 switch (flags & NOTE_TIMER_PRECMASK) { 731 case NOTE_SECONDS: 732 #ifdef __LP64__ 733 if (data > (SBT_MAX / SBT_1S)) 734 return (SBT_MAX); 735 #endif 736 return ((sbintime_t)data << 32); 737 case NOTE_MSECONDS: /* FALLTHROUGH */ 738 case 0: 739 if (data >= 1000) { 740 secs = data / 1000; 741 #ifdef __LP64__ 742 if (secs > (SBT_MAX / SBT_1S)) 743 return (SBT_MAX); 744 #endif 745 return (secs << 32 | MS_TO_SBT(data % 1000)); 746 } 747 return (MS_TO_SBT(data)); 748 case NOTE_USECONDS: 749 if (data >= 1000000) { 750 secs = data / 1000000; 751 #ifdef __LP64__ 752 if (secs > (SBT_MAX / SBT_1S)) 753 return (SBT_MAX); 754 #endif 755 return (secs << 32 | US_TO_SBT(data % 1000000)); 756 } 757 return (US_TO_SBT(data)); 758 case NOTE_NSECONDS: 759 if (data >= 1000000000) { 760 secs = data / 1000000000; 761 #ifdef __LP64__ 762 if (secs > (SBT_MAX / SBT_1S)) 763 return (SBT_MAX); 764 #endif 765 return (secs << 32 | NS_TO_SBT(data % 1000000000)); 766 } 767 return (NS_TO_SBT(data)); 768 default: 769 break; 770 } 771 return (-1); 772 } 773 774 struct kq_timer_cb_data { 775 struct callout c; 776 struct proc *p; 777 struct knote *kn; 778 int cpuid; 779 int flags; 780 TAILQ_ENTRY(kq_timer_cb_data) link; 781 sbintime_t next; /* next timer event fires at */ 782 sbintime_t to; /* precalculated timer period, 0 for abs */ 783 }; 784 785 #define KQ_TIMER_CB_ENQUEUED 0x01 786 787 static void 788 kqtimer_sched_callout(struct kq_timer_cb_data *kc) 789 { 790 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn, 791 kc->cpuid, C_ABSOLUTE); 792 } 793 794 void 795 kqtimer_proc_continue(struct proc *p) 796 { 797 struct kq_timer_cb_data *kc, *kc1; 798 struct bintime bt; 799 sbintime_t now; 800 801 PROC_LOCK_ASSERT(p, MA_OWNED); 802 803 getboottimebin(&bt); 804 now = bttosbt(bt); 805 806 TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) { 807 TAILQ_REMOVE(&p->p_kqtim_stop, kc, link); 808 kc->flags &= ~KQ_TIMER_CB_ENQUEUED; 809 if (kc->next <= now) 810 filt_timerexpire_l(kc->kn, true); 811 else 812 kqtimer_sched_callout(kc); 813 } 814 } 815 816 static void 817 filt_timerexpire_l(struct knote *kn, bool proc_locked) 818 { 819 struct kq_timer_cb_data *kc; 820 struct proc *p; 821 uint64_t delta; 822 sbintime_t now; 823 824 kc = kn->kn_ptr.p_v; 825 826 if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) { 827 kn->kn_data++; 828 KNOTE_ACTIVATE(kn, 0); 829 return; 830 } 831 832 now = sbinuptime(); 833 if (now >= kc->next) { 834 delta = (now - kc->next) / kc->to; 835 if (delta == 0) 836 delta = 1; 837 kn->kn_data += delta; 838 kc->next += delta * kc->to; 839 if (now >= kc->next) /* overflow */ 840 kc->next = now + kc->to; 841 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 842 } 843 844 /* 845 * Initial check for stopped kc->p is racy. It is fine to 846 * miss the set of the stop flags, at worst we would schedule 847 * one more callout. On the other hand, it is not fine to not 848 * schedule when we we missed clearing of the flags, we 849 * recheck them under the lock and observe consistent state. 850 */ 851 p = kc->p; 852 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 853 if (!proc_locked) 854 PROC_LOCK(p); 855 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 856 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) { 857 kc->flags |= KQ_TIMER_CB_ENQUEUED; 858 TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link); 859 } 860 if (!proc_locked) 861 PROC_UNLOCK(p); 862 return; 863 } 864 if (!proc_locked) 865 PROC_UNLOCK(p); 866 } 867 kqtimer_sched_callout(kc); 868 } 869 870 static void 871 filt_timerexpire(void *knx) 872 { 873 filt_timerexpire_l(knx, false); 874 } 875 876 /* 877 * data contains amount of time to sleep 878 */ 879 static int 880 filt_timervalidate(struct knote *kn, sbintime_t *to) 881 { 882 struct bintime bt; 883 sbintime_t sbt; 884 885 if (kn->kn_sdata < 0) 886 return (EINVAL); 887 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) 888 kn->kn_sdata = 1; 889 /* 890 * The only fflags values supported are the timer unit 891 * (precision) and the absolute time indicator. 892 */ 893 if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0) 894 return (EINVAL); 895 896 *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); 897 if (*to < 0) 898 return (EINVAL); 899 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 900 getboottimebin(&bt); 901 sbt = bttosbt(bt); 902 *to = MAX(0, *to - sbt); 903 } 904 return (0); 905 } 906 907 static int 908 filt_timerattach(struct knote *kn) 909 { 910 struct kq_timer_cb_data *kc; 911 sbintime_t to; 912 int error; 913 914 to = -1; 915 error = filt_timervalidate(kn, &to); 916 if (error != 0) 917 return (error); 918 KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 || 919 (kn->kn_sfflags & NOTE_ABSTIME) != 0, 920 ("%s: periodic timer has a calculated zero timeout", __func__)); 921 KASSERT(to >= 0, 922 ("%s: timer has a calculated negative timeout", __func__)); 923 924 if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) { 925 atomic_subtract_int(&kq_ncallouts, 1); 926 return (ENOMEM); 927 } 928 929 if ((kn->kn_sfflags & NOTE_ABSTIME) == 0) 930 kn->kn_flags |= EV_CLEAR; /* automatically set */ 931 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ 932 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); 933 kc->kn = kn; 934 kc->p = curproc; 935 kc->cpuid = PCPU_GET(cpuid); 936 kc->flags = 0; 937 callout_init(&kc->c, 1); 938 filt_timerstart(kn, to); 939 940 return (0); 941 } 942 943 static void 944 filt_timerstart(struct knote *kn, sbintime_t to) 945 { 946 struct kq_timer_cb_data *kc; 947 948 kc = kn->kn_ptr.p_v; 949 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 950 kc->next = to; 951 kc->to = 0; 952 } else { 953 kc->next = to + sbinuptime(); 954 kc->to = to; 955 } 956 kqtimer_sched_callout(kc); 957 } 958 959 static void 960 filt_timerdetach(struct knote *kn) 961 { 962 struct kq_timer_cb_data *kc; 963 unsigned int old __unused; 964 bool pending; 965 966 kc = kn->kn_ptr.p_v; 967 do { 968 callout_drain(&kc->c); 969 970 /* 971 * kqtimer_proc_continue() might have rescheduled this callout. 972 * Double-check, using the process mutex as an interlock. 973 */ 974 PROC_LOCK(kc->p); 975 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) { 976 kc->flags &= ~KQ_TIMER_CB_ENQUEUED; 977 TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link); 978 } 979 pending = callout_pending(&kc->c); 980 PROC_UNLOCK(kc->p); 981 } while (pending); 982 free(kc, M_KQUEUE); 983 old = atomic_fetchadd_int(&kq_ncallouts, -1); 984 KASSERT(old > 0, ("Number of callouts cannot become negative")); 985 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ 986 } 987 988 static void 989 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type) 990 { 991 struct kq_timer_cb_data *kc; 992 struct kqueue *kq; 993 sbintime_t to; 994 int error; 995 996 switch (type) { 997 case EVENT_REGISTER: 998 /* Handle re-added timers that update data/fflags */ 999 if (kev->flags & EV_ADD) { 1000 kc = kn->kn_ptr.p_v; 1001 1002 /* Drain any existing callout. */ 1003 callout_drain(&kc->c); 1004 1005 /* Throw away any existing undelivered record 1006 * of the timer expiration. This is done under 1007 * the presumption that if a process is 1008 * re-adding this timer with new parameters, 1009 * it is no longer interested in what may have 1010 * happened under the old parameters. If it is 1011 * interested, it can wait for the expiration, 1012 * delete the old timer definition, and then 1013 * add the new one. 1014 * 1015 * This has to be done while the kq is locked: 1016 * - if enqueued, dequeue 1017 * - make it no longer active 1018 * - clear the count of expiration events 1019 */ 1020 kq = kn->kn_kq; 1021 KQ_LOCK(kq); 1022 if (kn->kn_status & KN_QUEUED) 1023 knote_dequeue(kn); 1024 1025 kn->kn_status &= ~KN_ACTIVE; 1026 kn->kn_data = 0; 1027 KQ_UNLOCK(kq); 1028 1029 /* Reschedule timer based on new data/fflags */ 1030 kn->kn_sfflags = kev->fflags; 1031 kn->kn_sdata = kev->data; 1032 error = filt_timervalidate(kn, &to); 1033 if (error != 0) { 1034 kn->kn_flags |= EV_ERROR; 1035 kn->kn_data = error; 1036 } else 1037 filt_timerstart(kn, to); 1038 } 1039 break; 1040 1041 case EVENT_PROCESS: 1042 *kev = kn->kn_kevent; 1043 if (kn->kn_flags & EV_CLEAR) { 1044 kn->kn_data = 0; 1045 kn->kn_fflags = 0; 1046 } 1047 break; 1048 1049 default: 1050 panic("filt_timertouch() - invalid type (%ld)", type); 1051 break; 1052 } 1053 } 1054 1055 static int 1056 filt_timer(struct knote *kn, long hint) 1057 { 1058 1059 return (kn->kn_data != 0); 1060 } 1061 1062 static int 1063 filt_userattach(struct knote *kn) 1064 { 1065 1066 /* 1067 * EVFILT_USER knotes are not attached to anything in the kernel. 1068 */ 1069 kn->kn_hook = NULL; 1070 if (kn->kn_fflags & NOTE_TRIGGER) 1071 kn->kn_hookid = 1; 1072 else 1073 kn->kn_hookid = 0; 1074 return (0); 1075 } 1076 1077 static void 1078 filt_userdetach(__unused struct knote *kn) 1079 { 1080 1081 /* 1082 * EVFILT_USER knotes are not attached to anything in the kernel. 1083 */ 1084 } 1085 1086 static int 1087 filt_user(struct knote *kn, __unused long hint) 1088 { 1089 1090 return (kn->kn_hookid); 1091 } 1092 1093 static void 1094 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 1095 { 1096 u_int ffctrl; 1097 1098 switch (type) { 1099 case EVENT_REGISTER: 1100 if (kev->fflags & NOTE_TRIGGER) 1101 kn->kn_hookid = 1; 1102 1103 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 1104 kev->fflags &= NOTE_FFLAGSMASK; 1105 switch (ffctrl) { 1106 case NOTE_FFNOP: 1107 break; 1108 1109 case NOTE_FFAND: 1110 kn->kn_sfflags &= kev->fflags; 1111 break; 1112 1113 case NOTE_FFOR: 1114 kn->kn_sfflags |= kev->fflags; 1115 break; 1116 1117 case NOTE_FFCOPY: 1118 kn->kn_sfflags = kev->fflags; 1119 break; 1120 1121 default: 1122 /* XXX Return error? */ 1123 break; 1124 } 1125 kn->kn_sdata = kev->data; 1126 if (kev->flags & EV_CLEAR) { 1127 kn->kn_hookid = 0; 1128 kn->kn_data = 0; 1129 kn->kn_fflags = 0; 1130 } 1131 break; 1132 1133 case EVENT_PROCESS: 1134 *kev = kn->kn_kevent; 1135 kev->fflags = kn->kn_sfflags; 1136 kev->data = kn->kn_sdata; 1137 if (kn->kn_flags & EV_CLEAR) { 1138 kn->kn_hookid = 0; 1139 kn->kn_data = 0; 1140 kn->kn_fflags = 0; 1141 } 1142 break; 1143 1144 default: 1145 panic("filt_usertouch() - invalid type (%ld)", type); 1146 break; 1147 } 1148 } 1149 1150 int 1151 sys_kqueue(struct thread *td, struct kqueue_args *uap) 1152 { 1153 1154 return (kern_kqueue(td, 0, NULL)); 1155 } 1156 1157 int 1158 sys_kqueuex(struct thread *td, struct kqueuex_args *uap) 1159 { 1160 int flags; 1161 1162 if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0) 1163 return (EINVAL); 1164 flags = 0; 1165 if ((uap->flags & KQUEUE_CLOEXEC) != 0) 1166 flags |= O_CLOEXEC; 1167 return (kern_kqueue(td, flags, NULL)); 1168 } 1169 1170 static void 1171 kqueue_init(struct kqueue *kq) 1172 { 1173 1174 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); 1175 TAILQ_INIT(&kq->kq_head); 1176 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 1177 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 1178 } 1179 1180 int 1181 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) 1182 { 1183 struct filedesc *fdp; 1184 struct kqueue *kq; 1185 struct file *fp; 1186 struct ucred *cred; 1187 int fd, error; 1188 1189 fdp = td->td_proc->p_fd; 1190 cred = td->td_ucred; 1191 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) 1192 return (ENOMEM); 1193 1194 error = falloc_caps(td, &fp, &fd, flags, fcaps); 1195 if (error != 0) { 1196 chgkqcnt(cred->cr_ruidinfo, -1, 0); 1197 return (error); 1198 } 1199 1200 /* An extra reference on `fp' has been held for us by falloc(). */ 1201 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 1202 kqueue_init(kq); 1203 kq->kq_fdp = fdp; 1204 kq->kq_cred = crhold(cred); 1205 1206 FILEDESC_XLOCK(fdp); 1207 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 1208 FILEDESC_XUNLOCK(fdp); 1209 1210 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 1211 fdrop(fp, td); 1212 1213 td->td_retval[0] = fd; 1214 return (0); 1215 } 1216 1217 struct g_kevent_args { 1218 int fd; 1219 const void *changelist; 1220 int nchanges; 1221 void *eventlist; 1222 int nevents; 1223 const struct timespec *timeout; 1224 }; 1225 1226 int 1227 sys_kevent(struct thread *td, struct kevent_args *uap) 1228 { 1229 struct kevent_copyops k_ops = { 1230 .arg = uap, 1231 .k_copyout = kevent_copyout, 1232 .k_copyin = kevent_copyin, 1233 .kevent_size = sizeof(struct kevent), 1234 }; 1235 struct g_kevent_args gk_args = { 1236 .fd = uap->fd, 1237 .changelist = uap->changelist, 1238 .nchanges = uap->nchanges, 1239 .eventlist = uap->eventlist, 1240 .nevents = uap->nevents, 1241 .timeout = uap->timeout, 1242 }; 1243 1244 return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent")); 1245 } 1246 1247 static int 1248 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, 1249 struct kevent_copyops *k_ops, const char *struct_name) 1250 { 1251 struct timespec ts, *tsp; 1252 #ifdef KTRACE 1253 struct kevent *eventlist = uap->eventlist; 1254 #endif 1255 int error; 1256 1257 if (uap->timeout != NULL) { 1258 error = copyin(uap->timeout, &ts, sizeof(ts)); 1259 if (error) 1260 return (error); 1261 tsp = &ts; 1262 } else 1263 tsp = NULL; 1264 1265 #ifdef KTRACE 1266 if (KTRPOINT(td, KTR_STRUCT_ARRAY)) 1267 ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist, 1268 uap->nchanges, k_ops->kevent_size); 1269 #endif 1270 1271 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 1272 k_ops, tsp); 1273 1274 #ifdef KTRACE 1275 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) 1276 ktrstructarray(struct_name, UIO_USERSPACE, eventlist, 1277 td->td_retval[0], k_ops->kevent_size); 1278 #endif 1279 1280 return (error); 1281 } 1282 1283 /* 1284 * Copy 'count' items into the destination list pointed to by uap->eventlist. 1285 */ 1286 static int 1287 kevent_copyout(void *arg, struct kevent *kevp, int count) 1288 { 1289 struct kevent_args *uap; 1290 int error; 1291 1292 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1293 uap = (struct kevent_args *)arg; 1294 1295 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 1296 if (error == 0) 1297 uap->eventlist += count; 1298 return (error); 1299 } 1300 1301 /* 1302 * Copy 'count' items from the list pointed to by uap->changelist. 1303 */ 1304 static int 1305 kevent_copyin(void *arg, struct kevent *kevp, int count) 1306 { 1307 struct kevent_args *uap; 1308 int error; 1309 1310 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1311 uap = (struct kevent_args *)arg; 1312 1313 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 1314 if (error == 0) 1315 uap->changelist += count; 1316 return (error); 1317 } 1318 1319 #ifdef COMPAT_FREEBSD11 1320 static int 1321 kevent11_copyout(void *arg, struct kevent *kevp, int count) 1322 { 1323 struct freebsd11_kevent_args *uap; 1324 struct freebsd11_kevent kev11; 1325 int error, i; 1326 1327 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1328 uap = (struct freebsd11_kevent_args *)arg; 1329 1330 for (i = 0; i < count; i++) { 1331 kev11.ident = kevp->ident; 1332 kev11.filter = kevp->filter; 1333 kev11.flags = kevp->flags; 1334 kev11.fflags = kevp->fflags; 1335 kev11.data = kevp->data; 1336 kev11.udata = kevp->udata; 1337 error = copyout(&kev11, uap->eventlist, sizeof(kev11)); 1338 if (error != 0) 1339 break; 1340 uap->eventlist++; 1341 kevp++; 1342 } 1343 return (error); 1344 } 1345 1346 /* 1347 * Copy 'count' items from the list pointed to by uap->changelist. 1348 */ 1349 static int 1350 kevent11_copyin(void *arg, struct kevent *kevp, int count) 1351 { 1352 struct freebsd11_kevent_args *uap; 1353 struct freebsd11_kevent kev11; 1354 int error, i; 1355 1356 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1357 uap = (struct freebsd11_kevent_args *)arg; 1358 1359 for (i = 0; i < count; i++) { 1360 error = copyin(uap->changelist, &kev11, sizeof(kev11)); 1361 if (error != 0) 1362 break; 1363 kevp->ident = kev11.ident; 1364 kevp->filter = kev11.filter; 1365 kevp->flags = kev11.flags; 1366 kevp->fflags = kev11.fflags; 1367 kevp->data = (uintptr_t)kev11.data; 1368 kevp->udata = kev11.udata; 1369 bzero(&kevp->ext, sizeof(kevp->ext)); 1370 uap->changelist++; 1371 kevp++; 1372 } 1373 return (error); 1374 } 1375 1376 int 1377 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap) 1378 { 1379 struct kevent_copyops k_ops = { 1380 .arg = uap, 1381 .k_copyout = kevent11_copyout, 1382 .k_copyin = kevent11_copyin, 1383 .kevent_size = sizeof(struct freebsd11_kevent), 1384 }; 1385 struct g_kevent_args gk_args = { 1386 .fd = uap->fd, 1387 .changelist = uap->changelist, 1388 .nchanges = uap->nchanges, 1389 .eventlist = uap->eventlist, 1390 .nevents = uap->nevents, 1391 .timeout = uap->timeout, 1392 }; 1393 1394 return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent")); 1395 } 1396 #endif 1397 1398 int 1399 kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 1400 struct kevent_copyops *k_ops, const struct timespec *timeout) 1401 { 1402 cap_rights_t rights; 1403 struct file *fp; 1404 int error; 1405 1406 cap_rights_init_zero(&rights); 1407 if (nchanges > 0) 1408 cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE); 1409 if (nevents > 0) 1410 cap_rights_set_one(&rights, CAP_KQUEUE_EVENT); 1411 error = fget(td, fd, &rights, &fp); 1412 if (error != 0) 1413 return (error); 1414 1415 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); 1416 fdrop(fp, td); 1417 1418 return (error); 1419 } 1420 1421 static int 1422 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, 1423 struct kevent_copyops *k_ops, const struct timespec *timeout) 1424 { 1425 struct kevent keva[KQ_NEVENTS]; 1426 struct kevent *kevp, *changes; 1427 int i, n, nerrors, error; 1428 1429 if (nchanges < 0) 1430 return (EINVAL); 1431 1432 nerrors = 0; 1433 while (nchanges > 0) { 1434 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 1435 error = k_ops->k_copyin(k_ops->arg, keva, n); 1436 if (error) 1437 return (error); 1438 changes = keva; 1439 for (i = 0; i < n; i++) { 1440 kevp = &changes[i]; 1441 if (!kevp->filter) 1442 continue; 1443 kevp->flags &= ~EV_SYSFLAGS; 1444 error = kqueue_register(kq, kevp, td, M_WAITOK); 1445 if (error || (kevp->flags & EV_RECEIPT)) { 1446 if (nevents == 0) 1447 return (error); 1448 kevp->flags = EV_ERROR; 1449 kevp->data = error; 1450 (void)k_ops->k_copyout(k_ops->arg, kevp, 1); 1451 nevents--; 1452 nerrors++; 1453 } 1454 } 1455 nchanges -= n; 1456 } 1457 if (nerrors) { 1458 td->td_retval[0] = nerrors; 1459 return (0); 1460 } 1461 1462 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); 1463 } 1464 1465 int 1466 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, 1467 struct kevent_copyops *k_ops, const struct timespec *timeout) 1468 { 1469 struct kqueue *kq; 1470 int error; 1471 1472 error = kqueue_acquire(fp, &kq); 1473 if (error != 0) 1474 return (error); 1475 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); 1476 kqueue_release(kq, 0); 1477 return (error); 1478 } 1479 1480 /* 1481 * Performs a kevent() call on a temporarily created kqueue. This can be 1482 * used to perform one-shot polling, similar to poll() and select(). 1483 */ 1484 int 1485 kern_kevent_anonymous(struct thread *td, int nevents, 1486 struct kevent_copyops *k_ops) 1487 { 1488 struct kqueue kq = {}; 1489 int error; 1490 1491 kqueue_init(&kq); 1492 kq.kq_refcnt = 1; 1493 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); 1494 kqueue_drain(&kq, td); 1495 kqueue_destroy(&kq); 1496 return (error); 1497 } 1498 1499 int 1500 kqueue_add_filteropts(int filt, const struct filterops *filtops) 1501 { 1502 int error; 1503 1504 error = 0; 1505 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 1506 printf( 1507 "trying to add a filterop that is out of range: %d is beyond %d\n", 1508 ~filt, EVFILT_SYSCOUNT); 1509 return EINVAL; 1510 } 1511 mtx_lock(&filterops_lock); 1512 if (sysfilt_ops[~filt].for_fop != &null_filtops && 1513 sysfilt_ops[~filt].for_fop != NULL) 1514 error = EEXIST; 1515 else { 1516 sysfilt_ops[~filt].for_fop = filtops; 1517 sysfilt_ops[~filt].for_refcnt = 0; 1518 } 1519 mtx_unlock(&filterops_lock); 1520 1521 return (error); 1522 } 1523 1524 int 1525 kqueue_del_filteropts(int filt) 1526 { 1527 int error; 1528 1529 error = 0; 1530 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1531 return EINVAL; 1532 1533 mtx_lock(&filterops_lock); 1534 if (sysfilt_ops[~filt].for_fop == &null_filtops || 1535 sysfilt_ops[~filt].for_fop == NULL) 1536 error = EINVAL; 1537 else if (sysfilt_ops[~filt].for_refcnt != 0) 1538 error = EBUSY; 1539 else { 1540 sysfilt_ops[~filt].for_fop = &null_filtops; 1541 sysfilt_ops[~filt].for_refcnt = 0; 1542 } 1543 mtx_unlock(&filterops_lock); 1544 1545 return error; 1546 } 1547 1548 static const struct filterops * 1549 kqueue_fo_find(int filt) 1550 { 1551 1552 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1553 return NULL; 1554 1555 if (sysfilt_ops[~filt].for_nolock) 1556 return sysfilt_ops[~filt].for_fop; 1557 1558 mtx_lock(&filterops_lock); 1559 sysfilt_ops[~filt].for_refcnt++; 1560 if (sysfilt_ops[~filt].for_fop == NULL) 1561 sysfilt_ops[~filt].for_fop = &null_filtops; 1562 mtx_unlock(&filterops_lock); 1563 1564 return sysfilt_ops[~filt].for_fop; 1565 } 1566 1567 static void 1568 kqueue_fo_release(int filt) 1569 { 1570 1571 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1572 return; 1573 1574 if (sysfilt_ops[~filt].for_nolock) 1575 return; 1576 1577 mtx_lock(&filterops_lock); 1578 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 1579 ("filter object refcount not valid on release")); 1580 sysfilt_ops[~filt].for_refcnt--; 1581 mtx_unlock(&filterops_lock); 1582 } 1583 1584 /* 1585 * A ref to kq (obtained via kqueue_acquire) must be held. 1586 */ 1587 static int 1588 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, 1589 int mflag) 1590 { 1591 const struct filterops *fops; 1592 struct file *fp; 1593 struct knote *kn, *tkn; 1594 struct knlist *knl; 1595 int error, filt, event; 1596 int haskqglobal, filedesc_unlock; 1597 1598 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) 1599 return (EINVAL); 1600 1601 fp = NULL; 1602 kn = NULL; 1603 knl = NULL; 1604 error = 0; 1605 haskqglobal = 0; 1606 filedesc_unlock = 0; 1607 1608 filt = kev->filter; 1609 fops = kqueue_fo_find(filt); 1610 if (fops == NULL) 1611 return EINVAL; 1612 1613 if (kev->flags & EV_ADD) { 1614 /* Reject an invalid flag pair early */ 1615 if (kev->flags & EV_KEEPUDATA) { 1616 tkn = NULL; 1617 error = EINVAL; 1618 goto done; 1619 } 1620 1621 /* 1622 * Prevent waiting with locks. Non-sleepable 1623 * allocation failures are handled in the loop, only 1624 * if the spare knote appears to be actually required. 1625 */ 1626 tkn = knote_alloc(mflag); 1627 } else { 1628 tkn = NULL; 1629 } 1630 1631 findkn: 1632 if (fops->f_isfd) { 1633 KASSERT(td != NULL, ("td is NULL")); 1634 if (kev->ident > INT_MAX) 1635 error = EBADF; 1636 else 1637 error = fget(td, kev->ident, &cap_event_rights, &fp); 1638 if (error) 1639 goto done; 1640 1641 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 1642 kev->ident, M_NOWAIT) != 0) { 1643 /* try again */ 1644 fdrop(fp, td); 1645 fp = NULL; 1646 error = kqueue_expand(kq, fops, kev->ident, mflag); 1647 if (error) 1648 goto done; 1649 goto findkn; 1650 } 1651 1652 if (fp->f_type == DTYPE_KQUEUE) { 1653 /* 1654 * If we add some intelligence about what we are doing, 1655 * we should be able to support events on ourselves. 1656 * We need to know when we are doing this to prevent 1657 * getting both the knlist lock and the kq lock since 1658 * they are the same thing. 1659 */ 1660 if (fp->f_data == kq) { 1661 error = EINVAL; 1662 goto done; 1663 } 1664 1665 /* 1666 * Pre-lock the filedesc before the global 1667 * lock mutex, see the comment in 1668 * kqueue_close(). 1669 */ 1670 FILEDESC_XLOCK(td->td_proc->p_fd); 1671 filedesc_unlock = 1; 1672 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1673 } 1674 1675 KQ_LOCK(kq); 1676 if (kev->ident < kq->kq_knlistsize) { 1677 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1678 if (kev->filter == kn->kn_filter) 1679 break; 1680 } 1681 } else { 1682 if ((kev->flags & EV_ADD) == EV_ADD) { 1683 error = kqueue_expand(kq, fops, kev->ident, mflag); 1684 if (error != 0) 1685 goto done; 1686 } 1687 1688 KQ_LOCK(kq); 1689 1690 /* 1691 * If possible, find an existing knote to use for this kevent. 1692 */ 1693 if (kev->filter == EVFILT_PROC && 1694 (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { 1695 /* This is an internal creation of a process tracking 1696 * note. Don't attempt to coalesce this with an 1697 * existing note. 1698 */ 1699 ; 1700 } else if (kq->kq_knhashmask != 0) { 1701 struct klist *list; 1702 1703 list = &kq->kq_knhash[ 1704 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1705 SLIST_FOREACH(kn, list, kn_link) 1706 if (kev->ident == kn->kn_id && 1707 kev->filter == kn->kn_filter) 1708 break; 1709 } 1710 } 1711 1712 /* knote is in the process of changing, wait for it to stabilize. */ 1713 if (kn != NULL && kn_in_flux(kn)) { 1714 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1715 if (filedesc_unlock) { 1716 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1717 filedesc_unlock = 0; 1718 } 1719 kq->kq_state |= KQ_FLUXWAIT; 1720 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1721 if (fp != NULL) { 1722 fdrop(fp, td); 1723 fp = NULL; 1724 } 1725 goto findkn; 1726 } 1727 1728 /* 1729 * kn now contains the matching knote, or NULL if no match 1730 */ 1731 if (kn == NULL) { 1732 if (kev->flags & EV_ADD) { 1733 kn = tkn; 1734 tkn = NULL; 1735 if (kn == NULL) { 1736 KQ_UNLOCK(kq); 1737 error = ENOMEM; 1738 goto done; 1739 } 1740 kn->kn_fp = fp; 1741 kn->kn_kq = kq; 1742 kn->kn_fop = fops; 1743 /* 1744 * apply reference counts to knote structure, and 1745 * do not release it at the end of this routine. 1746 */ 1747 fops = NULL; 1748 fp = NULL; 1749 1750 kn->kn_sfflags = kev->fflags; 1751 kn->kn_sdata = kev->data; 1752 kev->fflags = 0; 1753 kev->data = 0; 1754 kn->kn_kevent = *kev; 1755 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1756 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); 1757 kn->kn_status = KN_DETACHED; 1758 if ((kev->flags & EV_DISABLE) != 0) 1759 kn->kn_status |= KN_DISABLED; 1760 kn_enter_flux(kn); 1761 1762 error = knote_attach(kn, kq); 1763 KQ_UNLOCK(kq); 1764 if (error != 0) { 1765 tkn = kn; 1766 goto done; 1767 } 1768 1769 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1770 knote_drop_detached(kn, td); 1771 goto done; 1772 } 1773 knl = kn_list_lock(kn); 1774 goto done_ev_add; 1775 } else { 1776 /* No matching knote and the EV_ADD flag is not set. */ 1777 KQ_UNLOCK(kq); 1778 error = ENOENT; 1779 goto done; 1780 } 1781 } 1782 1783 if (kev->flags & EV_DELETE) { 1784 kn_enter_flux(kn); 1785 KQ_UNLOCK(kq); 1786 knote_drop(kn, td); 1787 goto done; 1788 } 1789 1790 if (kev->flags & EV_FORCEONESHOT) { 1791 kn->kn_flags |= EV_ONESHOT; 1792 KNOTE_ACTIVATE(kn, 1); 1793 } 1794 1795 if ((kev->flags & EV_ENABLE) != 0) 1796 kn->kn_status &= ~KN_DISABLED; 1797 else if ((kev->flags & EV_DISABLE) != 0) 1798 kn->kn_status |= KN_DISABLED; 1799 1800 /* 1801 * The user may change some filter values after the initial EV_ADD, 1802 * but doing so will not reset any filter which has already been 1803 * triggered. 1804 */ 1805 kn->kn_status |= KN_SCAN; 1806 kn_enter_flux(kn); 1807 KQ_UNLOCK(kq); 1808 knl = kn_list_lock(kn); 1809 if ((kev->flags & EV_KEEPUDATA) == 0) 1810 kn->kn_kevent.udata = kev->udata; 1811 if (!fops->f_isfd && fops->f_touch != NULL) { 1812 fops->f_touch(kn, kev, EVENT_REGISTER); 1813 } else { 1814 kn->kn_sfflags = kev->fflags; 1815 kn->kn_sdata = kev->data; 1816 } 1817 1818 done_ev_add: 1819 /* 1820 * We can get here with kn->kn_knlist == NULL. This can happen when 1821 * the initial attach event decides that the event is "completed" 1822 * already, e.g., filt_procattach() is called on a zombie process. It 1823 * will call filt_proc() which will remove it from the list, and NULL 1824 * kn_knlist. 1825 * 1826 * KN_DISABLED will be stable while the knote is in flux, so the 1827 * unlocked read will not race with an update. 1828 */ 1829 if ((kn->kn_status & KN_DISABLED) == 0) 1830 event = kn->kn_fop->f_event(kn, 0); 1831 else 1832 event = 0; 1833 1834 KQ_LOCK(kq); 1835 if (event) 1836 kn->kn_status |= KN_ACTIVE; 1837 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == 1838 KN_ACTIVE) 1839 knote_enqueue(kn); 1840 kn->kn_status &= ~KN_SCAN; 1841 kn_leave_flux(kn); 1842 kn_list_unlock(knl); 1843 KQ_UNLOCK_FLUX(kq); 1844 1845 done: 1846 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1847 if (filedesc_unlock) 1848 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1849 if (fp != NULL) 1850 fdrop(fp, td); 1851 knote_free(tkn); 1852 if (fops != NULL) 1853 kqueue_fo_release(filt); 1854 return (error); 1855 } 1856 1857 static int 1858 kqueue_acquire(struct file *fp, struct kqueue **kqp) 1859 { 1860 int error; 1861 struct kqueue *kq; 1862 1863 error = 0; 1864 1865 kq = fp->f_data; 1866 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1867 return (EINVAL); 1868 *kqp = kq; 1869 KQ_LOCK(kq); 1870 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1871 KQ_UNLOCK(kq); 1872 return (EBADF); 1873 } 1874 kq->kq_refcnt++; 1875 KQ_UNLOCK(kq); 1876 1877 return error; 1878 } 1879 1880 static void 1881 kqueue_release(struct kqueue *kq, int locked) 1882 { 1883 if (locked) 1884 KQ_OWNED(kq); 1885 else 1886 KQ_LOCK(kq); 1887 kq->kq_refcnt--; 1888 if (kq->kq_refcnt == 1) 1889 wakeup(&kq->kq_refcnt); 1890 if (!locked) 1891 KQ_UNLOCK(kq); 1892 } 1893 1894 static void 1895 ast_kqueue(struct thread *td, int tda __unused) 1896 { 1897 taskqueue_quiesce(taskqueue_kqueue_ctx); 1898 } 1899 1900 static void 1901 kqueue_schedtask(struct kqueue *kq) 1902 { 1903 KQ_OWNED(kq); 1904 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1905 ("scheduling kqueue task while draining")); 1906 1907 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1908 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); 1909 kq->kq_state |= KQ_TASKSCHED; 1910 ast_sched(curthread, TDA_KQUEUE); 1911 } 1912 } 1913 1914 /* 1915 * Expand the kq to make sure we have storage for fops/ident pair. 1916 * 1917 * Return 0 on success (or no work necessary), return errno on failure. 1918 */ 1919 static int 1920 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident, 1921 int mflag) 1922 { 1923 struct klist *list, *tmp_knhash, *to_free; 1924 u_long tmp_knhashmask; 1925 int error, fd, size; 1926 1927 KQ_NOTOWNED(kq); 1928 1929 error = 0; 1930 to_free = NULL; 1931 if (fops->f_isfd) { 1932 fd = ident; 1933 if (kq->kq_knlistsize <= fd) { 1934 size = kq->kq_knlistsize; 1935 while (size <= fd) 1936 size += KQEXTENT; 1937 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1938 if (list == NULL) 1939 return ENOMEM; 1940 KQ_LOCK(kq); 1941 if ((kq->kq_state & KQ_CLOSING) != 0) { 1942 to_free = list; 1943 error = EBADF; 1944 } else if (kq->kq_knlistsize > fd) { 1945 to_free = list; 1946 } else { 1947 if (kq->kq_knlist != NULL) { 1948 bcopy(kq->kq_knlist, list, 1949 kq->kq_knlistsize * sizeof(*list)); 1950 to_free = kq->kq_knlist; 1951 kq->kq_knlist = NULL; 1952 } 1953 bzero((caddr_t)list + 1954 kq->kq_knlistsize * sizeof(*list), 1955 (size - kq->kq_knlistsize) * sizeof(*list)); 1956 kq->kq_knlistsize = size; 1957 kq->kq_knlist = list; 1958 } 1959 KQ_UNLOCK(kq); 1960 } 1961 } else { 1962 if (kq->kq_knhashmask == 0) { 1963 tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE, 1964 &tmp_knhashmask, (mflag & M_WAITOK) != 0 ? 1965 HASH_WAITOK : HASH_NOWAIT); 1966 if (tmp_knhash == NULL) 1967 return (ENOMEM); 1968 KQ_LOCK(kq); 1969 if ((kq->kq_state & KQ_CLOSING) != 0) { 1970 to_free = tmp_knhash; 1971 error = EBADF; 1972 } else if (kq->kq_knhashmask == 0) { 1973 kq->kq_knhash = tmp_knhash; 1974 kq->kq_knhashmask = tmp_knhashmask; 1975 } else { 1976 to_free = tmp_knhash; 1977 } 1978 KQ_UNLOCK(kq); 1979 } 1980 } 1981 free(to_free, M_KQUEUE); 1982 1983 KQ_NOTOWNED(kq); 1984 return (error); 1985 } 1986 1987 static void 1988 kqueue_task(void *arg, int pending) 1989 { 1990 struct kqueue *kq; 1991 int haskqglobal; 1992 1993 haskqglobal = 0; 1994 kq = arg; 1995 1996 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1997 KQ_LOCK(kq); 1998 1999 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 2000 2001 kq->kq_state &= ~KQ_TASKSCHED; 2002 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 2003 wakeup(&kq->kq_state); 2004 } 2005 KQ_UNLOCK(kq); 2006 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2007 } 2008 2009 /* 2010 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 2011 * We treat KN_MARKER knotes as if they are in flux. 2012 */ 2013 static int 2014 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 2015 const struct timespec *tsp, struct kevent *keva, struct thread *td) 2016 { 2017 struct kevent *kevp; 2018 struct knote *kn, *marker; 2019 struct knlist *knl; 2020 sbintime_t asbt, rsbt; 2021 int count, error, haskqglobal, influx, nkev, touch; 2022 2023 count = maxevents; 2024 nkev = 0; 2025 error = 0; 2026 haskqglobal = 0; 2027 2028 if (maxevents == 0) 2029 goto done_nl; 2030 if (maxevents < 0) { 2031 error = EINVAL; 2032 goto done_nl; 2033 } 2034 2035 rsbt = 0; 2036 if (tsp != NULL) { 2037 if (!timespecvalid_interval(tsp)) { 2038 error = EINVAL; 2039 goto done_nl; 2040 } 2041 if (timespecisset(tsp)) { 2042 if (tsp->tv_sec <= INT32_MAX) { 2043 rsbt = tstosbt(*tsp); 2044 if (TIMESEL(&asbt, rsbt)) 2045 asbt += tc_tick_sbt; 2046 if (asbt <= SBT_MAX - rsbt) 2047 asbt += rsbt; 2048 else 2049 asbt = 0; 2050 rsbt >>= tc_precexp; 2051 } else 2052 asbt = 0; 2053 } else 2054 asbt = -1; 2055 } else 2056 asbt = 0; 2057 marker = knote_alloc(M_WAITOK); 2058 marker->kn_status = KN_MARKER; 2059 KQ_LOCK(kq); 2060 2061 retry: 2062 kevp = keva; 2063 if (kq->kq_count == 0) { 2064 if (asbt == -1) { 2065 error = EWOULDBLOCK; 2066 } else { 2067 kq->kq_state |= KQ_SLEEP; 2068 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, 2069 "kqread", asbt, rsbt, C_ABSOLUTE); 2070 } 2071 if (error == 0) 2072 goto retry; 2073 /* don't restart after signals... */ 2074 if (error == ERESTART) 2075 error = EINTR; 2076 else if (error == EWOULDBLOCK) 2077 error = 0; 2078 goto done; 2079 } 2080 2081 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 2082 influx = 0; 2083 while (count) { 2084 KQ_OWNED(kq); 2085 kn = TAILQ_FIRST(&kq->kq_head); 2086 2087 if ((kn->kn_status == KN_MARKER && kn != marker) || 2088 kn_in_flux(kn)) { 2089 if (influx) { 2090 influx = 0; 2091 KQ_FLUX_WAKEUP(kq); 2092 } 2093 kq->kq_state |= KQ_FLUXWAIT; 2094 error = msleep(kq, &kq->kq_lock, PSOCK, 2095 "kqflxwt", 0); 2096 continue; 2097 } 2098 2099 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2100 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 2101 kn->kn_status &= ~KN_QUEUED; 2102 kq->kq_count--; 2103 continue; 2104 } 2105 if (kn == marker) { 2106 KQ_FLUX_WAKEUP(kq); 2107 if (count == maxevents) 2108 goto retry; 2109 goto done; 2110 } 2111 KASSERT(!kn_in_flux(kn), 2112 ("knote %p is unexpectedly in flux", kn)); 2113 2114 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 2115 kn->kn_status &= ~KN_QUEUED; 2116 kn_enter_flux(kn); 2117 kq->kq_count--; 2118 KQ_UNLOCK(kq); 2119 /* 2120 * We don't need to lock the list since we've 2121 * marked it as in flux. 2122 */ 2123 knote_drop(kn, td); 2124 KQ_LOCK(kq); 2125 continue; 2126 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 2127 kn->kn_status &= ~KN_QUEUED; 2128 kn_enter_flux(kn); 2129 kq->kq_count--; 2130 KQ_UNLOCK(kq); 2131 /* 2132 * We don't need to lock the list since we've 2133 * marked the knote as being in flux. 2134 */ 2135 *kevp = kn->kn_kevent; 2136 knote_drop(kn, td); 2137 KQ_LOCK(kq); 2138 kn = NULL; 2139 } else { 2140 kn->kn_status |= KN_SCAN; 2141 kn_enter_flux(kn); 2142 KQ_UNLOCK(kq); 2143 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 2144 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 2145 knl = kn_list_lock(kn); 2146 if (kn->kn_fop->f_event(kn, 0) == 0) { 2147 KQ_LOCK(kq); 2148 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2149 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | 2150 KN_SCAN); 2151 kn_leave_flux(kn); 2152 kq->kq_count--; 2153 kn_list_unlock(knl); 2154 influx = 1; 2155 continue; 2156 } 2157 touch = (!kn->kn_fop->f_isfd && 2158 kn->kn_fop->f_touch != NULL); 2159 if (touch) 2160 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 2161 else 2162 *kevp = kn->kn_kevent; 2163 KQ_LOCK(kq); 2164 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2165 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 2166 /* 2167 * Manually clear knotes who weren't 2168 * 'touch'ed. 2169 */ 2170 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 2171 kn->kn_data = 0; 2172 kn->kn_fflags = 0; 2173 } 2174 if (kn->kn_flags & EV_DISPATCH) 2175 kn->kn_status |= KN_DISABLED; 2176 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 2177 kq->kq_count--; 2178 } else 2179 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2180 2181 kn->kn_status &= ~KN_SCAN; 2182 kn_leave_flux(kn); 2183 kn_list_unlock(knl); 2184 influx = 1; 2185 } 2186 2187 /* we are returning a copy to the user */ 2188 kevp++; 2189 nkev++; 2190 count--; 2191 2192 if (nkev == KQ_NEVENTS) { 2193 influx = 0; 2194 KQ_UNLOCK_FLUX(kq); 2195 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 2196 nkev = 0; 2197 kevp = keva; 2198 KQ_LOCK(kq); 2199 if (error) 2200 break; 2201 } 2202 } 2203 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 2204 done: 2205 KQ_OWNED(kq); 2206 KQ_UNLOCK_FLUX(kq); 2207 knote_free(marker); 2208 done_nl: 2209 KQ_NOTOWNED(kq); 2210 if (nkev != 0) 2211 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 2212 td->td_retval[0] = maxevents - count; 2213 return (error); 2214 } 2215 2216 /*ARGSUSED*/ 2217 static int 2218 kqueue_ioctl(struct file *fp, u_long cmd, void *data, 2219 struct ucred *active_cred, struct thread *td) 2220 { 2221 /* 2222 * Enabling sigio causes two major problems: 2223 * 1) infinite recursion: 2224 * Synopsys: kevent is being used to track signals and have FIOASYNC 2225 * set. On receipt of a signal this will cause a kqueue to recurse 2226 * into itself over and over. Sending the sigio causes the kqueue 2227 * to become ready, which in turn posts sigio again, forever. 2228 * Solution: this can be solved by setting a flag in the kqueue that 2229 * we have a SIGIO in progress. 2230 * 2) locking problems: 2231 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 2232 * us above the proc and pgrp locks. 2233 * Solution: Post a signal using an async mechanism, being sure to 2234 * record a generation count in the delivery so that we do not deliver 2235 * a signal to the wrong process. 2236 * 2237 * Note, these two mechanisms are somewhat mutually exclusive! 2238 */ 2239 #if 0 2240 struct kqueue *kq; 2241 2242 kq = fp->f_data; 2243 switch (cmd) { 2244 case FIOASYNC: 2245 if (*(int *)data) { 2246 kq->kq_state |= KQ_ASYNC; 2247 } else { 2248 kq->kq_state &= ~KQ_ASYNC; 2249 } 2250 return (0); 2251 2252 case FIOSETOWN: 2253 return (fsetown(*(int *)data, &kq->kq_sigio)); 2254 2255 case FIOGETOWN: 2256 *(int *)data = fgetown(&kq->kq_sigio); 2257 return (0); 2258 } 2259 #endif 2260 2261 return (ENOTTY); 2262 } 2263 2264 /*ARGSUSED*/ 2265 static int 2266 kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 2267 struct thread *td) 2268 { 2269 struct kqueue *kq; 2270 int revents = 0; 2271 int error; 2272 2273 if ((error = kqueue_acquire(fp, &kq))) 2274 return POLLERR; 2275 2276 KQ_LOCK(kq); 2277 if (events & (POLLIN | POLLRDNORM)) { 2278 if (kq->kq_count) { 2279 revents |= events & (POLLIN | POLLRDNORM); 2280 } else { 2281 selrecord(td, &kq->kq_sel); 2282 if (SEL_WAITING(&kq->kq_sel)) 2283 kq->kq_state |= KQ_SEL; 2284 } 2285 } 2286 kqueue_release(kq, 1); 2287 KQ_UNLOCK(kq); 2288 return (revents); 2289 } 2290 2291 /*ARGSUSED*/ 2292 static int 2293 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred) 2294 { 2295 2296 bzero((void *)st, sizeof *st); 2297 /* 2298 * We no longer return kq_count because the unlocked value is useless. 2299 * If you spent all this time getting the count, why not spend your 2300 * syscall better by calling kevent? 2301 * 2302 * XXX - This is needed for libc_r. 2303 */ 2304 st->st_mode = S_IFIFO; 2305 return (0); 2306 } 2307 2308 static void 2309 kqueue_drain(struct kqueue *kq, struct thread *td) 2310 { 2311 struct knote *kn; 2312 int i; 2313 2314 KQ_LOCK(kq); 2315 2316 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 2317 ("kqueue already closing")); 2318 kq->kq_state |= KQ_CLOSING; 2319 if (kq->kq_refcnt > 1) 2320 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 2321 2322 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 2323 2324 KASSERT(knlist_empty(&kq->kq_sel.si_note), 2325 ("kqueue's knlist not empty")); 2326 2327 for (i = 0; i < kq->kq_knlistsize; i++) { 2328 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 2329 if (kn_in_flux(kn)) { 2330 kq->kq_state |= KQ_FLUXWAIT; 2331 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 2332 continue; 2333 } 2334 kn_enter_flux(kn); 2335 KQ_UNLOCK(kq); 2336 knote_drop(kn, td); 2337 KQ_LOCK(kq); 2338 } 2339 } 2340 if (kq->kq_knhashmask != 0) { 2341 for (i = 0; i <= kq->kq_knhashmask; i++) { 2342 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 2343 if (kn_in_flux(kn)) { 2344 kq->kq_state |= KQ_FLUXWAIT; 2345 msleep(kq, &kq->kq_lock, PSOCK, 2346 "kqclo2", 0); 2347 continue; 2348 } 2349 kn_enter_flux(kn); 2350 KQ_UNLOCK(kq); 2351 knote_drop(kn, td); 2352 KQ_LOCK(kq); 2353 } 2354 } 2355 } 2356 2357 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 2358 kq->kq_state |= KQ_TASKDRAIN; 2359 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 2360 } 2361 2362 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2363 selwakeuppri(&kq->kq_sel, PSOCK); 2364 if (!SEL_WAITING(&kq->kq_sel)) 2365 kq->kq_state &= ~KQ_SEL; 2366 } 2367 2368 KQ_UNLOCK(kq); 2369 } 2370 2371 static void 2372 kqueue_destroy(struct kqueue *kq) 2373 { 2374 2375 KASSERT(kq->kq_fdp == NULL, 2376 ("kqueue still attached to a file descriptor")); 2377 seldrain(&kq->kq_sel); 2378 knlist_destroy(&kq->kq_sel.si_note); 2379 mtx_destroy(&kq->kq_lock); 2380 2381 if (kq->kq_knhash != NULL) 2382 free(kq->kq_knhash, M_KQUEUE); 2383 if (kq->kq_knlist != NULL) 2384 free(kq->kq_knlist, M_KQUEUE); 2385 2386 funsetown(&kq->kq_sigio); 2387 } 2388 2389 /*ARGSUSED*/ 2390 static int 2391 kqueue_close(struct file *fp, struct thread *td) 2392 { 2393 struct kqueue *kq = fp->f_data; 2394 struct filedesc *fdp; 2395 int error; 2396 int filedesc_unlock; 2397 2398 if ((error = kqueue_acquire(fp, &kq))) 2399 return error; 2400 kqueue_drain(kq, td); 2401 2402 /* 2403 * We could be called due to the knote_drop() doing fdrop(), 2404 * called from kqueue_register(). In this case the global 2405 * lock is owned, and filedesc sx is locked before, to not 2406 * take the sleepable lock after non-sleepable. 2407 */ 2408 fdp = kq->kq_fdp; 2409 kq->kq_fdp = NULL; 2410 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 2411 FILEDESC_XLOCK(fdp); 2412 filedesc_unlock = 1; 2413 } else 2414 filedesc_unlock = 0; 2415 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 2416 if (filedesc_unlock) 2417 FILEDESC_XUNLOCK(fdp); 2418 2419 kqueue_destroy(kq); 2420 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); 2421 crfree(kq->kq_cred); 2422 free(kq, M_KQUEUE); 2423 fp->f_data = NULL; 2424 2425 return (0); 2426 } 2427 2428 static int 2429 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2430 { 2431 struct kqueue *kq = fp->f_data; 2432 2433 kif->kf_type = KF_TYPE_KQUEUE; 2434 kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq; 2435 kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count; 2436 kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state; 2437 return (0); 2438 } 2439 2440 static void 2441 kqueue_wakeup(struct kqueue *kq) 2442 { 2443 KQ_OWNED(kq); 2444 2445 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 2446 kq->kq_state &= ~KQ_SLEEP; 2447 wakeup(kq); 2448 } 2449 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2450 selwakeuppri(&kq->kq_sel, PSOCK); 2451 if (!SEL_WAITING(&kq->kq_sel)) 2452 kq->kq_state &= ~KQ_SEL; 2453 } 2454 if (!knlist_empty(&kq->kq_sel.si_note)) 2455 kqueue_schedtask(kq); 2456 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 2457 pgsigio(&kq->kq_sigio, SIGIO, 0); 2458 } 2459 } 2460 2461 /* 2462 * Walk down a list of knotes, activating them if their event has triggered. 2463 * 2464 * There is a possibility to optimize in the case of one kq watching another. 2465 * Instead of scheduling a task to wake it up, you could pass enough state 2466 * down the chain to make up the parent kqueue. Make this code functional 2467 * first. 2468 */ 2469 void 2470 knote(struct knlist *list, long hint, int lockflags) 2471 { 2472 struct kqueue *kq; 2473 struct knote *kn, *tkn; 2474 int error; 2475 2476 if (list == NULL) 2477 return; 2478 2479 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 2480 2481 if ((lockflags & KNF_LISTLOCKED) == 0) 2482 list->kl_lock(list->kl_lockarg); 2483 2484 /* 2485 * If we unlock the list lock (and enter influx), we can 2486 * eliminate the kqueue scheduling, but this will introduce 2487 * four lock/unlock's for each knote to test. Also, marker 2488 * would be needed to keep iteration position, since filters 2489 * or other threads could remove events. 2490 */ 2491 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { 2492 kq = kn->kn_kq; 2493 KQ_LOCK(kq); 2494 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 2495 /* 2496 * Do not process the influx notes, except for 2497 * the influx coming from the kq unlock in the 2498 * kqueue_scan(). In the later case, we do 2499 * not interfere with the scan, since the code 2500 * fragment in kqueue_scan() locks the knlist, 2501 * and cannot proceed until we finished. 2502 */ 2503 KQ_UNLOCK(kq); 2504 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 2505 kn_enter_flux(kn); 2506 KQ_UNLOCK(kq); 2507 error = kn->kn_fop->f_event(kn, hint); 2508 KQ_LOCK(kq); 2509 kn_leave_flux(kn); 2510 if (error) 2511 KNOTE_ACTIVATE(kn, 1); 2512 KQ_UNLOCK_FLUX(kq); 2513 } else { 2514 if (kn->kn_fop->f_event(kn, hint)) 2515 KNOTE_ACTIVATE(kn, 1); 2516 KQ_UNLOCK(kq); 2517 } 2518 } 2519 if ((lockflags & KNF_LISTLOCKED) == 0) 2520 list->kl_unlock(list->kl_lockarg); 2521 } 2522 2523 /* 2524 * add a knote to a knlist 2525 */ 2526 void 2527 knlist_add(struct knlist *knl, struct knote *kn, int islocked) 2528 { 2529 2530 KNL_ASSERT_LOCK(knl, islocked); 2531 KQ_NOTOWNED(kn->kn_kq); 2532 KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn)); 2533 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2534 ("knote %p was not detached", kn)); 2535 if (!islocked) 2536 knl->kl_lock(knl->kl_lockarg); 2537 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 2538 if (!islocked) 2539 knl->kl_unlock(knl->kl_lockarg); 2540 KQ_LOCK(kn->kn_kq); 2541 kn->kn_knlist = knl; 2542 kn->kn_status &= ~KN_DETACHED; 2543 KQ_UNLOCK(kn->kn_kq); 2544 } 2545 2546 static void 2547 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, 2548 int kqislocked) 2549 { 2550 2551 KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked")); 2552 KNL_ASSERT_LOCK(knl, knlislocked); 2553 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 2554 KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn)); 2555 KASSERT((kn->kn_status & KN_DETACHED) == 0, 2556 ("knote %p was already detached", kn)); 2557 if (!knlislocked) 2558 knl->kl_lock(knl->kl_lockarg); 2559 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 2560 kn->kn_knlist = NULL; 2561 if (!knlislocked) 2562 kn_list_unlock(knl); 2563 if (!kqislocked) 2564 KQ_LOCK(kn->kn_kq); 2565 kn->kn_status |= KN_DETACHED; 2566 if (!kqislocked) 2567 KQ_UNLOCK(kn->kn_kq); 2568 } 2569 2570 /* 2571 * remove knote from the specified knlist 2572 */ 2573 void 2574 knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 2575 { 2576 2577 knlist_remove_kq(knl, kn, islocked, 0); 2578 } 2579 2580 int 2581 knlist_empty(struct knlist *knl) 2582 { 2583 2584 KNL_ASSERT_LOCKED(knl); 2585 return (SLIST_EMPTY(&knl->kl_list)); 2586 } 2587 2588 static struct mtx knlist_lock; 2589 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 2590 MTX_DEF); 2591 static void knlist_mtx_lock(void *arg); 2592 static void knlist_mtx_unlock(void *arg); 2593 2594 static void 2595 knlist_mtx_lock(void *arg) 2596 { 2597 2598 mtx_lock((struct mtx *)arg); 2599 } 2600 2601 static void 2602 knlist_mtx_unlock(void *arg) 2603 { 2604 2605 mtx_unlock((struct mtx *)arg); 2606 } 2607 2608 static void 2609 knlist_mtx_assert_lock(void *arg, int what) 2610 { 2611 2612 if (what == LA_LOCKED) 2613 mtx_assert((struct mtx *)arg, MA_OWNED); 2614 else 2615 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 2616 } 2617 2618 void 2619 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 2620 void (*kl_unlock)(void *), 2621 void (*kl_assert_lock)(void *, int)) 2622 { 2623 2624 if (lock == NULL) 2625 knl->kl_lockarg = &knlist_lock; 2626 else 2627 knl->kl_lockarg = lock; 2628 2629 if (kl_lock == NULL) 2630 knl->kl_lock = knlist_mtx_lock; 2631 else 2632 knl->kl_lock = kl_lock; 2633 if (kl_unlock == NULL) 2634 knl->kl_unlock = knlist_mtx_unlock; 2635 else 2636 knl->kl_unlock = kl_unlock; 2637 if (kl_assert_lock == NULL) 2638 knl->kl_assert_lock = knlist_mtx_assert_lock; 2639 else 2640 knl->kl_assert_lock = kl_assert_lock; 2641 2642 knl->kl_autodestroy = 0; 2643 SLIST_INIT(&knl->kl_list); 2644 } 2645 2646 void 2647 knlist_init_mtx(struct knlist *knl, struct mtx *lock) 2648 { 2649 2650 knlist_init(knl, lock, NULL, NULL, NULL); 2651 } 2652 2653 struct knlist * 2654 knlist_alloc(struct mtx *lock) 2655 { 2656 struct knlist *knl; 2657 2658 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); 2659 knlist_init_mtx(knl, lock); 2660 return (knl); 2661 } 2662 2663 void 2664 knlist_destroy(struct knlist *knl) 2665 { 2666 2667 KASSERT(KNLIST_EMPTY(knl), 2668 ("destroying knlist %p with knotes on it", knl)); 2669 } 2670 2671 void 2672 knlist_detach(struct knlist *knl) 2673 { 2674 2675 KNL_ASSERT_LOCKED(knl); 2676 knl->kl_autodestroy = 1; 2677 if (knlist_empty(knl)) { 2678 knlist_destroy(knl); 2679 free(knl, M_KQUEUE); 2680 } 2681 } 2682 2683 /* 2684 * Even if we are locked, we may need to drop the lock to allow any influx 2685 * knotes time to "settle". 2686 */ 2687 void 2688 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2689 { 2690 struct knote *kn, *kn2; 2691 struct kqueue *kq; 2692 2693 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); 2694 if (islocked) 2695 KNL_ASSERT_LOCKED(knl); 2696 else { 2697 KNL_ASSERT_UNLOCKED(knl); 2698 again: /* need to reacquire lock since we have dropped it */ 2699 knl->kl_lock(knl->kl_lockarg); 2700 } 2701 2702 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2703 kq = kn->kn_kq; 2704 KQ_LOCK(kq); 2705 if (kn_in_flux(kn)) { 2706 KQ_UNLOCK(kq); 2707 continue; 2708 } 2709 knlist_remove_kq(knl, kn, 1, 1); 2710 if (killkn) { 2711 kn_enter_flux(kn); 2712 KQ_UNLOCK(kq); 2713 knote_drop_detached(kn, td); 2714 } else { 2715 /* Make sure cleared knotes disappear soon */ 2716 kn->kn_flags |= EV_EOF | EV_ONESHOT; 2717 KQ_UNLOCK(kq); 2718 } 2719 kq = NULL; 2720 } 2721 2722 if (!SLIST_EMPTY(&knl->kl_list)) { 2723 /* there are still in flux knotes remaining */ 2724 kn = SLIST_FIRST(&knl->kl_list); 2725 kq = kn->kn_kq; 2726 KQ_LOCK(kq); 2727 KASSERT(kn_in_flux(kn), ("knote removed w/o list lock")); 2728 knl->kl_unlock(knl->kl_lockarg); 2729 kq->kq_state |= KQ_FLUXWAIT; 2730 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2731 kq = NULL; 2732 goto again; 2733 } 2734 2735 if (islocked) 2736 KNL_ASSERT_LOCKED(knl); 2737 else { 2738 knl->kl_unlock(knl->kl_lockarg); 2739 KNL_ASSERT_UNLOCKED(knl); 2740 } 2741 } 2742 2743 /* 2744 * Remove all knotes referencing a specified fd must be called with FILEDESC 2745 * lock. This prevents a race where a new fd comes along and occupies the 2746 * entry and we attach a knote to the fd. 2747 */ 2748 void 2749 knote_fdclose(struct thread *td, int fd) 2750 { 2751 struct filedesc *fdp = td->td_proc->p_fd; 2752 struct kqueue *kq; 2753 struct knote *kn; 2754 int influx; 2755 2756 FILEDESC_XLOCK_ASSERT(fdp); 2757 2758 /* 2759 * We shouldn't have to worry about new kevents appearing on fd 2760 * since filedesc is locked. 2761 */ 2762 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2763 KQ_LOCK(kq); 2764 2765 again: 2766 influx = 0; 2767 while (kq->kq_knlistsize > fd && 2768 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2769 if (kn_in_flux(kn)) { 2770 /* someone else might be waiting on our knote */ 2771 if (influx) 2772 wakeup(kq); 2773 kq->kq_state |= KQ_FLUXWAIT; 2774 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2775 goto again; 2776 } 2777 kn_enter_flux(kn); 2778 KQ_UNLOCK(kq); 2779 influx = 1; 2780 knote_drop(kn, td); 2781 KQ_LOCK(kq); 2782 } 2783 KQ_UNLOCK_FLUX(kq); 2784 } 2785 } 2786 2787 static int 2788 knote_attach(struct knote *kn, struct kqueue *kq) 2789 { 2790 struct klist *list; 2791 2792 KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn)); 2793 KQ_OWNED(kq); 2794 2795 if ((kq->kq_state & KQ_CLOSING) != 0) 2796 return (EBADF); 2797 if (kn->kn_fop->f_isfd) { 2798 if (kn->kn_id >= kq->kq_knlistsize) 2799 return (ENOMEM); 2800 list = &kq->kq_knlist[kn->kn_id]; 2801 } else { 2802 if (kq->kq_knhash == NULL) 2803 return (ENOMEM); 2804 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2805 } 2806 SLIST_INSERT_HEAD(list, kn, kn_link); 2807 return (0); 2808 } 2809 2810 static void 2811 knote_drop(struct knote *kn, struct thread *td) 2812 { 2813 2814 if ((kn->kn_status & KN_DETACHED) == 0) 2815 kn->kn_fop->f_detach(kn); 2816 knote_drop_detached(kn, td); 2817 } 2818 2819 static void 2820 knote_drop_detached(struct knote *kn, struct thread *td) 2821 { 2822 struct kqueue *kq; 2823 struct klist *list; 2824 2825 kq = kn->kn_kq; 2826 2827 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2828 ("knote %p still attached", kn)); 2829 KQ_NOTOWNED(kq); 2830 2831 KQ_LOCK(kq); 2832 for (;;) { 2833 KASSERT(kn->kn_influx >= 1, 2834 ("knote_drop called on %p with influx %d", 2835 kn, kn->kn_influx)); 2836 if (kn->kn_influx == 1) 2837 break; 2838 kq->kq_state |= KQ_FLUXWAIT; 2839 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2840 } 2841 2842 if (kn->kn_fop->f_isfd) 2843 list = &kq->kq_knlist[kn->kn_id]; 2844 else 2845 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2846 2847 if (!SLIST_EMPTY(list)) 2848 SLIST_REMOVE(list, kn, knote, kn_link); 2849 if (kn->kn_status & KN_QUEUED) 2850 knote_dequeue(kn); 2851 KQ_UNLOCK_FLUX(kq); 2852 2853 if (kn->kn_fop->f_isfd) { 2854 fdrop(kn->kn_fp, td); 2855 kn->kn_fp = NULL; 2856 } 2857 kqueue_fo_release(kn->kn_kevent.filter); 2858 kn->kn_fop = NULL; 2859 knote_free(kn); 2860 } 2861 2862 static void 2863 knote_enqueue(struct knote *kn) 2864 { 2865 struct kqueue *kq = kn->kn_kq; 2866 2867 KQ_OWNED(kn->kn_kq); 2868 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2869 2870 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2871 kn->kn_status |= KN_QUEUED; 2872 kq->kq_count++; 2873 kqueue_wakeup(kq); 2874 } 2875 2876 static void 2877 knote_dequeue(struct knote *kn) 2878 { 2879 struct kqueue *kq = kn->kn_kq; 2880 2881 KQ_OWNED(kn->kn_kq); 2882 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2883 2884 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2885 kn->kn_status &= ~KN_QUEUED; 2886 kq->kq_count--; 2887 } 2888 2889 static void 2890 knote_init(void) 2891 { 2892 2893 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2894 NULL, NULL, UMA_ALIGN_PTR, 0); 2895 ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue); 2896 prison0.pr_klist = knlist_alloc(&prison0.pr_mtx); 2897 } 2898 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2899 2900 static struct knote * 2901 knote_alloc(int mflag) 2902 { 2903 2904 return (uma_zalloc(knote_zone, mflag | M_ZERO)); 2905 } 2906 2907 static void 2908 knote_free(struct knote *kn) 2909 { 2910 2911 uma_zfree(knote_zone, kn); 2912 } 2913 2914 /* 2915 * Register the kev w/ the kq specified by fd. 2916 */ 2917 int 2918 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag) 2919 { 2920 struct kqueue *kq; 2921 struct file *fp; 2922 cap_rights_t rights; 2923 int error; 2924 2925 error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE), 2926 &fp); 2927 if (error != 0) 2928 return (error); 2929 if ((error = kqueue_acquire(fp, &kq)) != 0) 2930 goto noacquire; 2931 2932 error = kqueue_register(kq, kev, td, mflag); 2933 kqueue_release(kq, 0); 2934 2935 noacquire: 2936 fdrop(fp, td); 2937 return (error); 2938 } 2939 2940 struct knote_status_export_bit { 2941 int kn_status_bit; 2942 int knt_status_bit; 2943 }; 2944 2945 #define ST(name) \ 2946 { .kn_status_bit = KN_##name, .knt_status_bit = KNOTE_STATUS_##name } 2947 static const struct knote_status_export_bit knote_status_export_bits[] = { 2948 ST(ACTIVE), 2949 ST(QUEUED), 2950 ST(DISABLED), 2951 ST(DETACHED), 2952 ST(KQUEUE), 2953 }; 2954 #undef ST 2955 2956 static int 2957 knote_status_export(int kn_status) 2958 { 2959 const struct knote_status_export_bit *b; 2960 unsigned i; 2961 int res; 2962 2963 res = 0; 2964 for (i = 0; i < nitems(knote_status_export_bits); i++) { 2965 b = &knote_status_export_bits[i]; 2966 if ((kn_status & b->kn_status_bit) != 0) 2967 res |= b->knt_status_bit; 2968 } 2969 return (res); 2970 } 2971 2972 static int 2973 kern_proc_kqueue_report_one(struct sbuf *s, struct proc *p, 2974 int kq_fd, struct kqueue *kq, struct knote *kn, bool compat32 __unused) 2975 { 2976 struct kinfo_knote kin; 2977 #ifdef COMPAT_FREEBSD32 2978 struct kinfo_knote32 kin32; 2979 #endif 2980 int error; 2981 2982 if (kn->kn_status == KN_MARKER) 2983 return (0); 2984 2985 memset(&kin, 0, sizeof(kin)); 2986 kin.knt_kq_fd = kq_fd; 2987 memcpy(&kin.knt_event, &kn->kn_kevent, sizeof(struct kevent)); 2988 kin.knt_status = knote_status_export(kn->kn_status); 2989 kn_enter_flux(kn); 2990 KQ_UNLOCK_FLUX(kq); 2991 if (kn->kn_fop->f_userdump != NULL) 2992 (void)kn->kn_fop->f_userdump(p, kn, &kin); 2993 #ifdef COMPAT_FREEBSD32 2994 if (compat32) { 2995 freebsd32_kinfo_knote_to_32(&kin, &kin32); 2996 error = sbuf_bcat(s, &kin32, sizeof(kin32)); 2997 } else 2998 #endif 2999 error = sbuf_bcat(s, &kin, sizeof(kin)); 3000 KQ_LOCK(kq); 3001 kn_leave_flux(kn); 3002 return (error); 3003 } 3004 3005 static int 3006 kern_proc_kqueue_report(struct sbuf *s, struct proc *p, int kq_fd, 3007 struct kqueue *kq, bool compat32) 3008 { 3009 struct knote *kn; 3010 int error, i; 3011 3012 error = 0; 3013 KQ_LOCK(kq); 3014 for (i = 0; i < kq->kq_knlistsize; i++) { 3015 SLIST_FOREACH(kn, &kq->kq_knlist[i], kn_link) { 3016 error = kern_proc_kqueue_report_one(s, p, kq_fd, 3017 kq, kn, compat32); 3018 if (error != 0) 3019 goto out; 3020 } 3021 } 3022 if (kq->kq_knhashmask == 0) 3023 goto out; 3024 for (i = 0; i <= kq->kq_knhashmask; i++) { 3025 SLIST_FOREACH(kn, &kq->kq_knhash[i], kn_link) { 3026 error = kern_proc_kqueue_report_one(s, p, kq_fd, 3027 kq, kn, compat32); 3028 if (error != 0) 3029 goto out; 3030 } 3031 } 3032 out: 3033 KQ_UNLOCK_FLUX(kq); 3034 return (error); 3035 } 3036 3037 struct kern_proc_kqueues_out1_cb_args { 3038 struct sbuf *s; 3039 bool compat32; 3040 }; 3041 3042 static int 3043 kern_proc_kqueues_out1_cb(struct proc *p, int fd, struct file *fp, void *arg) 3044 { 3045 struct kqueue *kq; 3046 struct kern_proc_kqueues_out1_cb_args *a; 3047 3048 if (fp->f_type != DTYPE_KQUEUE) 3049 return (0); 3050 a = arg; 3051 kq = fp->f_data; 3052 return (kern_proc_kqueue_report(a->s, p, fd, kq, a->compat32)); 3053 } 3054 3055 static int 3056 kern_proc_kqueues_out1(struct thread *td, struct proc *p, struct sbuf *s, 3057 bool compat32) 3058 { 3059 struct kern_proc_kqueues_out1_cb_args a; 3060 3061 a.s = s; 3062 a.compat32 = compat32; 3063 return (fget_remote_foreach(td, p, kern_proc_kqueues_out1_cb, &a)); 3064 } 3065 3066 int 3067 kern_proc_kqueues_out(struct proc *p, struct sbuf *sb, size_t maxlen, 3068 bool compat32) 3069 { 3070 struct sbuf *s, sm; 3071 size_t sb_len; 3072 int error; 3073 3074 if (maxlen == -1 || maxlen == 0) 3075 sb_len = 128; 3076 else 3077 sb_len = maxlen; 3078 s = sbuf_new(&sm, NULL, sb_len, maxlen == -1 ? SBUF_AUTOEXTEND : 3079 SBUF_FIXEDLEN); 3080 error = kern_proc_kqueues_out1(curthread, p, s, compat32); 3081 sbuf_finish(s); 3082 if (error == 0) { 3083 sbuf_bcat(sb, sbuf_data(s), MIN(sbuf_len(s), maxlen == -1 ? 3084 SIZE_T_MAX : maxlen)); 3085 } 3086 sbuf_delete(s); 3087 return (error); 3088 } 3089 3090 static int 3091 sysctl_kern_proc_kqueue_one(struct thread *td, struct sbuf *s, struct proc *p, 3092 int kq_fd, bool compat32) 3093 { 3094 struct file *fp; 3095 struct kqueue *kq; 3096 int error; 3097 3098 error = fget_remote(td, p, kq_fd, &fp); 3099 if (error == 0) { 3100 if (fp->f_type != DTYPE_KQUEUE) { 3101 error = EINVAL; 3102 } else { 3103 kq = fp->f_data; 3104 error = kern_proc_kqueue_report(s, p, kq_fd, kq, 3105 compat32); 3106 } 3107 fdrop(fp, td); 3108 } 3109 return (error); 3110 } 3111 3112 static int 3113 sysctl_kern_proc_kqueue(SYSCTL_HANDLER_ARGS) 3114 { 3115 struct thread *td; 3116 struct proc *p; 3117 struct sbuf *s, sm; 3118 int error, error1, *name; 3119 bool compat32; 3120 3121 name = (int *)arg1; 3122 if ((u_int)arg2 > 2 || (u_int)arg2 == 0) 3123 return (EINVAL); 3124 3125 error = pget((pid_t)name[0], PGET_HOLD | PGET_CANDEBUG, &p); 3126 if (error != 0) 3127 return (error); 3128 3129 td = curthread; 3130 #ifdef COMPAT_FREEBSD32 3131 compat32 = SV_CURPROC_FLAG(SV_ILP32); 3132 #else 3133 compat32 = false; 3134 #endif 3135 3136 s = sbuf_new_for_sysctl(&sm, NULL, 0, req); 3137 if (s == NULL) { 3138 error = ENOMEM; 3139 goto out; 3140 } 3141 sbuf_clear_flags(s, SBUF_INCLUDENUL); 3142 3143 if ((u_int)arg2 == 1) { 3144 error = kern_proc_kqueues_out1(td, p, s, compat32); 3145 } else { 3146 error = sysctl_kern_proc_kqueue_one(td, s, p, 3147 name[1] /* kq_fd */, compat32); 3148 } 3149 3150 error1 = sbuf_finish(s); 3151 if (error == 0) 3152 error = error1; 3153 sbuf_delete(s); 3154 3155 out: 3156 PRELE(p); 3157 return (error); 3158 } 3159 3160 static SYSCTL_NODE(_kern_proc, KERN_PROC_KQUEUE, kq, 3161 CTLFLAG_RD | CTLFLAG_MPSAFE, 3162 sysctl_kern_proc_kqueue, "KQueue events"); 3163